Cross-Domain Social Rumor-Propagation Model Based on Transfer Learning.

Area of Science:

• Computer Science
• Social Network Analysis
• Computational Social Science

Background:

• Rumors exhibit varied text characteristics across domains but share emotional similarities.
• Scarce data in specific rumor-topic domains hinders accurate propagation prediction.
• Understanding user emotions and network structures is crucial for rumor analysis.

Purpose of the Study:

• To propose a novel cross-domain rumor-propagation model using transfer learning.
• To address the challenge of data scarcity in rumor-topic domains.
• To enhance the accuracy of rumor propagation prediction across different domains.

Main Methods:

• Developed User-Retweet-Rumor2vec (URR2vec) for latent feature extraction and representation learning.
• Employed a deep learning model for text sentiment analysis to capture user emotional cognition.
• Proposed a Text-Sentiment Analysis-Graph Convolutional Network (TSA-GCN) for prediction, enhanced with domain adaptation.

Main Results:

• URR2vec effectively represents user-retweet-rumor relationships and topic information.
• The sentiment analysis model identifies emotional correlations among users during rumor spread.
• The TSA-GCN model, with domain adaptation, achieves accurate cross-domain rumor propagation prediction.

Conclusions:

• The proposed transfer learning model effectively overcomes data scarcity issues in rumor propagation.
• Domain adaptation techniques improve prediction accuracy by migrating parameters and graph structures.
• This approach offers a robust solution for analyzing and predicting rumor dynamics across diverse topics.